2020-05-05 14:52:02

by Jia-Ju Bai

[permalink] [raw]
Subject: [PATCH v2] fs: jfs: fix a possible data race in metapage_writepage()

The functions metapage_writepage() and lmPostGC() can be concurrently
executed in the following call contexts:

Thread1:
metapage_writepage()

Thread2:
lbmIODone()
lmPostGC()

In metapage_writepage():
if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))

In lmPostGC():
spin_lock_irqsave(&log->gclock, flags);
...
log->cflag &= ~logGC_PAGEOUT
...
spin_unlock_irqrestore(&log->gclock, flags);

The memory addresses of mp->log->cflag and log->cflag can be identical,
and thus a data race can occur.
This data race was found by our concurrency fuzzer.

Thus use the spin lock "mp->log->gclock" for the assignment of
the data structure member "log->cflag" to a local variable
in this function implementation.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
v2:
* Change the description.
Thank Markus Elfring for good advice.

fs/jfs/jfs_metapage.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/fs/jfs/jfs_metapage.c b/fs/jfs/jfs_metapage.c
index a2f5338a5ea1..026c11b2572d 100644
--- a/fs/jfs/jfs_metapage.c
+++ b/fs/jfs/jfs_metapage.c
@@ -351,6 +351,7 @@ static int metapage_writepage(struct page *page, struct writeback_control *wbc)
unsigned long bio_offset = 0;
int offset;
int bad_blocks = 0;
+ uint cflag;

page_start = (sector_t)page->index <<
(PAGE_SHIFT - inode->i_blkbits);
@@ -370,8 +371,14 @@ static int metapage_writepage(struct page *page, struct writeback_control *wbc)
* Make sure this page isn't blocked indefinitely.
* If the journal isn't undergoing I/O, push it
*/
- if (mp->log && !(mp->log->cflag & logGC_PAGEOUT))
- jfs_flush_journal(mp->log, 0);
+
+ if (mp->log) {
+ spin_lock_irq(&mp->log->gclock);
+ cflag = mp->log->cflag;
+ spin_unlock_irq(&mp->log->gclock);
+ if (!(cflag & logGC_PAGEOUT))
+ jfs_flush_journal(mp->log, 0);
+ }
continue;
}

--
2.17.1